Sloud Video Content Monitoring System

The market for online video content is expanding rapidly. Parks Associates forecast, the Internet video services will reach more than $7 billion in revenue by 2010. Such rapid expansion created demand for automated monitoring of digital video content on web-servers and in the large databases.

Answering market demand, Sloud has started development of a video content monitoring system. Initially this system will permit indexing content of video files and video monitoring, including finding of video duplicates and copyright infringing videos. The system is targeted at video sites with user-uploaded content.

The first version of the system utilizes Sloud's well established MRM technology for indexing and searching video files. MRM technology was developed for indexing and identification of digital audio records, primarily music.

MRM is based on acoustic (audio) fingerprinting. MRM is applied to video content by indexing an audio track extracted from the video. The indexed audio is used for identification of the video file as a whole.

An advantage of such "video-on-audio" identification is high reliability of identification even in case of distorted (clipped, re-compressed, re-sampled) or otherwise changed video data. The algorithm is resistant to a low or even moderate amount of noise in the audio. This alleviated the known problem of other video fingerprinting systems when simple distortion of video, such as changed aspect ration, makes algorithms unreliable.

Sloud technology does not possess flaws inherently present in products based on hash functions and checksums. The identification based on hash functions, particularly cryptographic hash functions, fails when even one bit in data file is changed. MRM naturally ignores substantial parts of the file: it ignores the video content and parts of audio spectrum. When the changes affect the audio part used by MRM, the changes have to be substantial before MRM fails to detect similarity. Furthermore, the "same-not same" threshold can be adjusted for specific needs (requires re-indexing).

The following diagram shows the main components of the system:

• DCMM - Duplicate Checking and Monitoring Module
• EI - Editor's Interface
• IDFS - Indexing and Duplicates Finding Service
• UI - User's Interface
• VDB - Video Database
• VF - Video Fingerprint
• VFDB - Video Fingerprints Database

Sloud has an agreement with a publicly traded company to deploy the prototype by the end of July 2007. The prototype provides video content monitoring system for the company's user-uploaded video content.

By the end of 2007 Sloud is planning to release technology for monitoring video based on identifying visual component of the video stream (video-fingerprinting in addition to acoustic fingerprinting of the initial version). The technology will be resistant to cropping, aspect ratio changes, and even some tilting.

We conducted a patent search and believe the algorithms are original and are not infringing on any patents.